Orthogonal variance decomposition based feature selection

Existing feature selection methods fail to properly account for interactions between features. In this paper, we attempt remedy issue by using orthogonal variance decomposition evaluate The orthogonality of the allows us directly calculate total contribution each output variance. As a result, obtain an efficient and technically sound algorithm which takes into interactions. proposed has low computational complexity compared other used in literature. Numerical experiments demonstrate that our method accurately identifies relevant features improves accuracy numerical models.